Pete, Our network is currently offering DiffServ QoS to customers. It's a new service, but so far it is working well and has been very well received by our customers. I'd be happy to answer any questions about what we do or why offline but here are some "talking points" of our policy. 1. We established a network wide delay budget and then put together some software that monitors network latency. 2. We control the classification feature. We classify all our customer traffic for them with Cisco class based policing at the CPE router. 3. We use Cisco's LLQ or MDRR queuing only on our links from distribution to the network edge. On our high speed links (622Mb and above) we don't queue. There are two schools of thought on this, but we currently subscribe to the theory that just letting FIFO happen is faster than applying custom queues in the core. YMMV with very fast routers, ASICs, and all that. Hope that helps! Mike -----Original Message----- From: Pete Kruckenberg [mailto:pete@kruckenberg.com] Sent: Monday, February 03, 2003 1:04 PM To: nanog@merit.edu Subject: IP QoS case-studies I've found there's no shortage of advice and theory about the viability of IP QoS (DiffServ) in a large wide-area (converged) network. I have not had much luck with finding documentation about experiences implementing and operating such a beast. Presumably that's yet another (silent) confirmation that It Doesn't Work or There's a Better/Easier Way. Nevertheless, I'd still like to find anyone who has tried (successfully or not) to converge (ie VoIP/H.323/data) a high-speed (~ 1Gb/s) IP network and use IP QoS for what it is sold to do. White paper/presentation references or off-line conversation would be appreciated. Pete.